MITCHELL J. LOVETT, RENANA PERES, and RON SHACHAR*
Brands and word of mouth (WOM) are cornerstones of the marketing
field, and yet their relationship has received relatively little attention. This
study aims to enhance understanding of brand characteristics as
antecedents of WOM by executing a comprehensive empirical analysis.
For this purpose, the authors constructed a unique data set on online
and offline WOM and characteristics for more than 600 of the most
talked-about U.S. brands. To guide this empirical analysis, they present a
theoretical framework arguing that consumers spread WOM on brands
as a result of social, emotional, and functional drivers. Using these
drivers, the authors identify a set of 13 brand characteristics that
stimulate WOM, including three (level of differentiation, excitement, and
complexity) that have not been studied to date as WOM antecedents.
The authors find that whereas the social and functional drivers are the
most important for online WOM, the emotional driver is the most
important for offline WOM. These results provide an insightful
perspective on WOM and have meaningful managerial implications for
brand management and investment in WOM campaigns.
Keywords: word of mouth, brands, complexity, differentiation, esteem,
online, offline
On Brands and Word of Mouth
Do differentiated brands get more or less word of mouth
(WOM) than others? Does the level of differentiation matter
in terms of WOM? What about the degree of a brand’s complexity or its level of excitement? Extant research does not
answer these questions. More broadly, although brands and
WOM are cornerstones of the marketing field, the relationship
between them has received relatively little attention. Although
the literature has explored many aspects of the impact (e.g.,
Chevalier and Mayzlin 2006), dynamics (e.g., Yang et al.
2012), and social networking dimensions (e.g., Goldenberg
et al. 2006; Katona, Zubcsek, and Sarvary 2011) of WOM,
understanding of brand characteristics as antecedents to WOM
is surprisingly limited. That is, the two broad literature streams
on branding and WOM are largely distinct. Nevertheless, the
role of brand characteristics in WOM is not only critical but
also highly relevant for marketing scholars and practitioners
for a variety of reasons, such as to create “talkable brands” and
maximize the impact of branding activities (Rosen 2002;
Word of Mouth Marketing Association [WOMMA] 2011).
This study aims to enhance the field’s understanding of
brand characteristics as antecedents of WOM by executing
a comprehensive empirical analysis that examines the brand
characteristics–WOM relationships for numerous brands.
Specifically, we collected data on the 697 most talked-about
U.S. national brands from 16 categories (e.g., food, media
and entertainment, cars, financial services, sports), which
include product and service brands, corporate brands, and
product-specific brands. Our analysis is comprehensive not
only because of the large number of brands but also because
we incorporate both online and offline measures of WOM,
*Mitchell J. Lovett is Assistant Professor of Marketing, Simon Business
School, University of Rochester (e-mail: [email protected].
edu). Renana Peres is Lecturer (Assistant Professor) of Marketing, School
of Business Administration, Hebrew University of Jerusalem (e-mail:
[email protected]). Ron Shachar is Dean and Professor, Arison School of
Business, Interdisciplinary Center Herzliya (e-mail: [email protected]).
The authors thank their industry collaborators—Brad Fay from the Keller
Fay Group, the NM Incite team, and Ed Lebar from Y&R’s Brand Asset
Valuator—for sharing their data. In addition, the authors thank Kristin
Luck and the Decipher Inc. team for programming and managing the survey, Eitan Muller and Barak Libai for fruitful discussions, and the participants of the Marketing Science conference and the Yale Customer Insights
conference. They are extremely grateful to their research assistants—at
Wharton: Christina Andrews, Linda Wang, Chris Webber-Deonauth, Derric Bath, Grace Choi, Rachel Amalo, Yan Yan, Niels Mayrargue, Nathan
Pamart, and Fangdan Chen; at Hebrew University of Jerusalem: Yair
Cohen, Dafna Presler, Oshri Weiss, Liron Zaretzky, Anna Proviz, Tal
Tamir, and Haneen Matar. This research was supported by the Marketing
Science Institute, the Wharton Customer Analytics Initiative, the Israel
Internet Association, Kmart International Center for Marketing and Retailing at the Hebrew University of Jerusalem, the Israel Science Foundation,
and the Marketing Department at the Wharton School. Last, the authors
thank the review team for their comments and insights. Jeffrey Inman
served as associate editor for this article.
© 2013, American Marketing Association
ISSN: 0022-2437 (print), 1547-7193 (electronic)
427
Journal of Marketing Research
Vol. L (August 2013), 427–444
428
whereas existing research has typically relied on one or
the other. Furthermore, previous scholars studying the
antecedents of WOM have focused on only one or two
brand characteristics; in our study, we evaluate the role of a
broad set of brand characteristics.
To guide this empirical analysis, we begin by developing
a theoretical framework that identifies brand characteristics
that are relevant for WOM. This framework, whose fundamentals are consumers and the factors that stimulate them
to engage in WOM, argues that consumers spread WOM for
brands as a result of three drivers: social, emotional, and
functional. The social driver relates to social signaling (i.e.,
expressing uniqueness, self-enhancement, and a desire to
socialize), the emotional driver is related to emotion sharing, and the functional driver is related to the need to obtain
and the tendency to provide information. Understanding
these drivers and the needs associated with them helps us
identify specific brand characteristics that play a role in
stimulating WOM. Consider, for example, the social driver—
specifically, the need to express uniqueness: it is easier to
signal uniqueness through a highly differentiated brand than
an undifferentiated brand. Consequently, we argue that a
brand with a higher degree of differentiation is likely to
have greater WOM. Notably, the potential role of differentiation on WOM has not yet been studied. Two additional
characteristics that are novel to our study are a brand’s level
of excitement and its complexity.
Another layer of our analysis involves the heterogeneity of
WOM across channels of communication. Specifically, the
purpose and nature of WOM differ between offline conversations and online brand mentions. Whereas offline communication typically occurs in a one-to-one context and carries nonverbal clues, online communication is typically written and
in a one-to-many context (i.e., read by a great number of
people). Accordingly, we expect that the drivers’ impact will
differ between the two channels of communications as well.
Our empirical analysis rests on a comprehensive data set
that includes the 697 most talked-about U.S. national
brands. For each of these brands, we compiled data on
offline and online WOM and, in line with our theoretical
framework, on their characteristics. We gathered the brand
characteristics data from (1) a survey we conducted on a
representative U.S. sample with 4,769 respondents (on characteristics such as complexity and excitement) and (2) the
proprietary Y&R data based on its Brand Asset Valuator
panel (on characteristics such as differentiation). The data
on the offline WOM are from the Keller Fay group (Keller
2007) and include a weekly measure of the offline WOM
(i.e., face-to-face and telephone conversations) for more
than 1,000 brands mentioned from January 2007 to August
2010. The online data are from the former Nielsen and
McKinsey NM Incite’s tool, and they include a daily measure of the online WOM (i.e., blogs, user forums, and Twitter messages) for each of these brands between 2008 and
2010.
Our analysis of this cross-sectional data not only provides
empirical support for the relationship between brand characteristics and WOM but also demonstrates that each of the
drivers identified in our theoretical framework (social, emotional, and functional) is relevant and significant in this
process. Furthermore, each of the characteristics this framework introduces to the WOM discussion (differentiation,
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
complexity, and excitement) has a significant relationship
with WOM. For example, we find that brands that are
highly differentiated from others (and thus enable consumers to express their uniqueness) have, as expected, more
WOM. Notably, this effect is much stronger in the online
setting than in offline conversations.
Indeed, the results also reveal insightful differences
between online and offline WOM at the brand characteristic
level. In some cases, characteristics have a significant effect
in one setting but not in the other (e.g., age of brand). These
discrepancies at the brand characteristic level are indicative
of differences with respect to the importance of the three
overall drivers. We find that whereas the social and functional drivers are the most important for online WOM, the
emotional driver is the most important for offline WOM.
These results paint a unique picture of WOM. Offline conversations, which typically occur in one-on-one settings, are
inherently more personal and intimate and thus enable people to share emotions such as excitement and satisfaction.
Online WOM, which usually involves “broadcasting” to
many people (e.g., Twitter), may be more appropriate for
social signaling (e.g., emphasizing uniqueness).
Our work not only reveals new findings; it also has managerial implications. Brand managers could leverage our
results to help diagnose their brands’ WOM performance.
For example, our model could be used to identify brands
that, given their characteristics, underperform in terms of
WOM in the sense that their actual WOM is lower than the
level predicted by the model. Such a gap might be due to
various sources, one of which is that the brand has not lived
up to its WOM potential, which might suggest a reexamination of the firm’s WOM strategy. To assist managers in diagnosing their brands’ WOM performance, we created a
model-based Descriptive Decision Support System (DDSS)
in Microsoft Excel (Power and Sharda 2009).
THEORETICAL FRAMEWORK
This section introduces our theoretical framework, which
we then use to identify brand characteristics that are relevant for WOM. We begin with the most fundamental elements: consumers and the factors that stimulate them to
engage in WOM. Building on previous research, we argue
that consumers spread brand WOM for three fundamental
purposes: social, emotional, and functional. The main social
driver is the desire to send signals to others about one’s
expertise, uniqueness, or social status; the emotional driver
is the need to share positive or negative feelings about
brands to balance emotional arousal; and the functional
driver motivates people to provide and supply information.
There may be additional drivers, but drawing on previous
research, we consider these three the major ones.
We use the term “theoretical framework” rather than
“theory” or “model” to reflect its role adequately in this
study. This study’s main contribution rests on the data and
findings, and the central role of the theoretical framework is
to guide the empirical analysis by identifying and organizing brand characteristics that might be relevant for WOM as
well as by suggesting possible interpretation for the results.
Next, we discuss in more detail the relevant literature and
the three fundamental drivers and identify their relevant
brand characteristics. For clarity, the brand characteristics
that are included in our model are in italics.
On Brands and Word of Mouth
The Social Driver
Self-enhancement. A compelling social motive to engage
in WOM is self-enhancement. Wojnicki and Godes (2011)
show that consumers strategically use WOM to signal or
enhance their perceived expertise. To achieve this purpose,
positive WOM is more effective than negative, because
experts are expected to identify high-quality products better
than novices. Thus, consistent with prior evidence (Amblee
and Bui 2008), it is expected that the higher the esteem or
quality associated with the brand, the more likely consumers are to engage in WOM about it. Another aspect of
self-enhancement is status signaling. People use their purchases to signal their social status to others, either to their
own social group or to other groups (Han, Nunes, and Drèze
2010) such that luxury goods signal a high social status
(Veblen 1994 [1899]). We suggest that consumers can signal a high social status not only by purchasing but also by
talking about luxury goods. Therefore, we hypothesize that
brands that are perceived as premium will generate a higher
level of WOM than what people refer to as value brands.
Expressing uniqueness. Previous studies have demonstrated that consumers use consumption and possessions to
express their uniqueness or their group identity (e.g., Berger
and Heath 2007), but surprisingly, researches have ignored
the possibility of employing WOM for this purpose.1 We
suggest that consumers can also express their uniqueness by
talking about brands. Furthermore, we posit that some
brands are better suited to express uniqueness than others.
Specifically, brands that are highly differentiated from others more easily enable consumers to project a unique identity or membership in a group. Therefore, we hypothesize
that the higher the degree of a brand’s differentiation, the
more likely it is to generate WOM.
Desire to converse. The basic human desire to socialize,
and thus converse, with others (Rosen 2002; Rubin, Perese,
and Barbato 1988) can lead to WOM. Berger and Schwartz
(2011) demonstrate empirically that a brand’s visibility
facilitates people’s ability to use it in a conversation. Thus,
our model will account for the brand’s visibility or observability. Another attribute that may make a brand suitable for
conversation is whether it is relevant in the lives of many
people. For example, an indie band is less likely to be conversation material than a mainstream band.2 Consequently,
we expect that as brands become more relevant to more
people, they are more likely to spark conversation.
The Emotional Driver
Consuming or thinking about a brand can evoke emotions
that people might like to share with others (Heath, Bell, and
Sternberg 2001; Nardi et al. 2004; Peters and Kashima 2007)
to express or ease emotional arousal (Berger and Milkman
1Previous studies that have examined the interaction between WOM and
expression of uniqueness have had very different foci. Ho and Dempsey
(2010) show that people who stand up to others report that they are more
likely to forward online content. Che, Lurie, and Weiss (2011) demonstrate
that people who want to communicate their expertise have a higher tendency to reply to online requests for advice. Cheema and Kaikati (2010)
find that people with high need for uniqueness refrain from engaging in
WOM to keep others away from “their” products.
2This should be true even when we take into account that a fan of an
indie band is likely to have similar friends. Even in the fan’s personal life,
there are probably many people who are not likely to be fans of indie bands.
429
2012). Previous studies have focused on the role of one
emotion in this context: satisfaction. They provide evidence
that brands that evoke both very high (Roberts 2004) and very
low (Richins 1983) satisfaction levels receive higher levels
of WOM than brands with moderate levels of satisfaction.
Notably, the role of a brand’s excitement (one of the five
brand personality traits introduced in Aaker 1997) has been
overlooked in this context. However, excitement is certainly
a stimulating emotion that can be expressed through conversation, and thus, it is reasonable to expect that the higher the
brand’s excitement factor, the more likely people are to
engage in WOM about it.
The Functional Driver
People may exchange useful and practical information
through conversation, and brands are often the subject of
that information exchange. In any such exchange, there is a
person who needs the information and one who provides it.
We proceed by discussing these two sides.
Information demand. Previous studies have suggested
that consumers’ need for information is especially high for
new brands because the uncertainty associated with them is
higher (for a review, see Peres, Muller, and Mahajan 2010).
Furthermore, existing evidence indirectly suggests that
WOM decreases over the life of a brand (Godes and Mayzlin 2004). Accordingly, we include the age of the brand
and directly test whether WOM is greater for newer brands.
Whereas previous studies have focused on the newness
of the brand as a source of uncertainty, we suggest that
another characteristic (heretofore ignored by previous studies on WOM) might be in play: the brand’s complexity, or
the difficulty of obtaining and comprehending information
about it. We hypothesize that the greater the complexity, the
greater the brand’s WOM. Note that although prior research
has ignored the relationship between this characteristic and
WOM, it has been discussed in the context of diffusion of
innovations (Rogers 1995).
The demand for information might also depend on the
type of good, be it an experience, search, or credence good
(Anand and Shachar 2011; Mudambi and Schuff 2010). In
this context, WOM can be useful for exploring intangible
attributes of experience goods (e.g., ambience in a restaurant) and for keeping up to date on observable attributes of
search goods (e.g., new service plans with AT&T). Whether
search goods, experience goods, or credence goods stimulate more WOM, however, is an open empirical question.
Information supply. Previous studies have identified
motives to provide information (e.g., altruism, reciprocity).
For a consumer to provide information and engage in conversation about a brand, it must be familiar to him or her.
Thus, we hypothesize that a higher level of familiarity (Sundaram and Webster 1999) or knowledge about a brand will
be associated with more WOM.
Hybrid Characteristics
Two additional brand characteristics discussed in previous studies, involvement (Dichter 1966; Sundaram, Mitra,
and Webster 1998) and perceived risk (Lutz and Reilly
1974; Sundaram, Mitra, and Webster 1998), do not fit well
into a single driver. Involvement can be both functional and
emotional: it can be functional because people are likely to
seek more information about high-involvement products,
430
and it can be emotional because some commonly used
scales of involvement include items such as “means a lot to
me” (Zaichkowsky 1985), which reflect emotions that people may feel the need to share. Similarly, perceived risk can
also be mapped into both the functional and the emotional
drivers. Rogers (1995) discusses three aspects of risk: the
actual performance of the brand, the extra expenses that
might be incurred, and the social embarrassment the brand
might cause. Although each of these risks might motivate
consumers to seek information to resolve them, they might
also induce anxiety that consumers may want to express, as
explained by Sundaram, Mitra, and Webster (1998), who
focus on this emotional aspect of risk.
As we discuss subsequently, some of our empirical analysis
is intended to evaluate the relative importance of the three
fundamental drivers. The classification of two characteristics
as both functional and emotional complicates this analysis.
To address this issue, we execute the analysis both with and
without the hybrid characteristics to demonstrate robustness.
Figure 1 illustrates our theoretical framework, including
the three fundamental drivers—social, emotional, and functional—and the associated brand characteristics. We propose that these brand characteristics affect the level of
WOM. In the following subsection, we describe our measures and data collection procedures for these brand characteristics and for WOM on both online and offline channels.
Online Versus Offline
Thus far, we have discussed WOM without distinguishing
offline conversations and online mentions. However, it is
Figure 1
THEORETICAL FRAMEWORK: MATCHING WOM DRIVERS,
BRAND CHARACTERISTICS AND WOM
Social
Quality
Differentiation
Premium/Value
Relevance
Visibility
Emotional
Excitement
Satisfaction
Perceived Risk*
Involvement*
Functional
Age
Complexity
Type of Good
Knowledge
Perceived Risk*
Involvement*
*Hybrid characteristics.
Word-of-Mouth
Mentions
(Online or offline)
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
reasonable to assume that the purpose and nature of WOM
differ between the two environments. First, offline meetings
are more intimate and personal than online interactions
because, unlike online interactions in which a person “broadcasts” a message to many (e.g., Facebook, Twitter), offline
conversations frequently occur in a personal one-on-one
setting (Hoffman and Novak 1996; Morris and Ogan 1996).
Second, in offline meetings (especially those that are faceto-face), the communication extends beyond spoken words.
For example, people can use tone, facial expressions, and
body language to convey thought. Third, offline interactions
are more interactive (or “synchronous,” in communications
terminology), in the sense that the other conversation parties are expected to respond, usually immediately (Morris
and Ogan 1996). In contrast, online channels such as blogs,
user forums, and Twitter are, in many cases, unidirectional
and asynchronous, with no immediate response (if any).
As a result of such differences, it is reasonable to expect
that the roles of the drivers and characteristics differ
between the two communication media. Consider the three
characteristics that have not previously been studied as
WOM antecedents: differentiation, excitement, and complexity. As we discussed, differentiation enables consumers
to express their uniqueness. Because in most online interactions, (1) the format is one of broadcasting to many (Morris and Ogan 1996) and (2) interactions are more likely to
take place with unfamiliar people (Walther 1996), the tendency to express personal aspects in general, and uniqueness in particular, should be greater. Furthermore, the ability to express uniqueness nonverbally is greater in an offline
setting than in online interactions (Illouz 2007, chap. 3). For
example, when a person wears a shirt from a highly differentiated brand, he or she need not mention it in an offline
conversation. As a result of these two factors, differentiation might be expected to play a more important role in
online versus offline interactions.
In contrast, the relative role of excitement is likely to be
stronger in offline meetings than in online interactions due
to the intimate and personal nature of the former. In addition, when excited, people might seek immediate responses
and feedback from their conversation partners; therefore,
they might prefer to use the more interactive and synchronous offline medium. As a result, we expect excitement to
be stronger offline than online.
Finally, the impact of complexity is likely to be stronger
in offline meetings than in online interactions due to the
interactivity of the offline medium as well as the ability to
ask and answer questions, which helps people comprehend
complex brands. This stands in contrast to online communications, which are not only asynchronous but also sometimes limited in length (e.g., Twitter, user forums) and thus
restrict thorough discussion.
Although (1) we do not offer a clear theory on this matter
and (2) the focus of our study is not to understand these differences, per se, we study the effect of brand characteristics
on online and offline WOM separately to avoid misspecification of the model. We also aim to provide some initial
empirical insights on this issue.
DATA
To study the role of brand characteristics in stimulating
WOM, we used several sources to build a comprehensive
On Brands and Word of Mouth
data set containing information about WOM as well as
brand characteristics for 697 major U.S. national brands
spanning 16 broad product categories (for the full list of
brands and categories as well as the data set’s construction,
see Web Appendix A at www.marketingpower.com/jmr_
webappendix).3 The categories are beauty products, beverages, cars, children’s products, clothing products, department stores, financial services, food and dining, health
products, home design and decoration, household products,
media and entertainment, sports and hobbies, technology
products and stores, telecommunication, and travel services.
The brands’ heterogeneity is high, including both corporate
and product brands. These include consumer brands (e.g.,
Coca-Cola, Dove), service brands (e.g., Expedia, Charles
Schwab, Burger King), sports teams (e.g., the Boston
Celtics), and television shows (e.g., CSI: Crime Scene
Investigation). For each brand, we collected data on WOM,
brand characteristics, and relevant control variables. Figure
2 describes our complete set of data sources, which we
describe in detail in the following subsections.
WOM Data
We group the variety of channels through which WOM can
be distributed and consumed into two main categories: offline
channels, such as face-to-face and telephone conversations,
and online channels, such as blogs, e-mails, user reviews, vir-
3We compiled this list on the basis of our WOM data to capture the most
talked-about brands in the United States between 2007 and 2010.
Figure 2
THE LIST OF DATA SOURCES
Explanatory variables
Dependent variables
Complexity, Visibility,
Involvement, Excitement,
Familiarity, Perceived Risk
The Keller Fay Group
Decipher Inc.
Brand Asset Valuator
by Y&R
Differentiation, Relevance,
Esteem, Knowledge,
Usage, Satisfaction
Interbrand
Brand Equity—is it part of
the top 100?
Secondary Data
Collection
Age, Type of Good,
Product/Service, Premium/
Value, Internet Brand
Offline Word of Mouth
NM Incite
Online Word of Mouth
431
tual social networks, user forums, and microblogs (e.g., Twitter). We collect data on the overall number of brand mentions
during the study’s time period for both channel categories
and conduct our analysis on the two categories separately.
Offline WOM. The Keller Fay Group’s TalkTrack project
is the industry’s most accepted measure of offline WOM
(e.g., it is used by WOMMA). This is a diary-style survey
of a representative sample of the U.S. population. Every
week, 700 different respondents are asked to conduct a 24hour diary in which they document WOM incidents, including every face-to-face or phone conversation they have in
which a brand is mentioned. Then, they list the brands mentioned in the conversation. Note that a list of brands is not
provided to respondents (i.e., they can mention any brand).
For each brand, we aggregated the number of mentions
between January 2007 and August 2010 and included both
telephone and face-to-face conversations. The average number of mentions in our data is 805, and the brand with the
highest number (15,038) is Coca-Cola.
Online WOM. The source for the online WOM data is the
tool used by Nielsen McKinsey’s NM Incite (formerly BuzzMetrics), a search engine that has conducted daily searches
through blogs, discussion groups, and microblogs since July
2008 and processes all available posts for each of these
sources.4 As with the offline data, we aggregated the data
across time (July 2008 to March 2010) and online sources.
The average number of online mentions in our data is
approximately 430,000, and the brand with the greatest
number of mentions (14,579,172) is Google.
Table 1 displays the top ten brands online and offline.
Note that these include both product brands, such as iPhone
and Xbox 360, and corporate brands, such as Sony and
AT&T. Only one brand, Ford, appears in both lists, illustrating the differences between these two WOM channels.
Table 2 presents the distribution of mentions across the 16
categories. For each category, it shows the number of
brands and the average number of mentions per brand for
offline and online settings.
The way we obtained the brand mentions differs between
the two channels. For the offline data, we used a sample of
4Operating this search engine requires the user to build queries that
include the brand and related words to retrieve the relevant information and
distinguish the brand from unrelated mentions of the same name (e.g.,
some brand names are also everyday words, such as the television show
House or Gap clothing stores). In addition, the tool excludes automatic
reposts, such as retweets. We also manually checked a large sample of
posts and found that more than 95% seemed to be user generated.
Table 1
TOP TEN MOST MENTIONED BRANDS OFFLINE AND ONLINE
Order
1
2
3
4
5
6
7
8
9
10
Offline
Coca-Cola
Verizon
Pepsi
Wal-Mart
Ford
AT&T
McDonald’s
Dell
Sony
Chevrolet
Online
Google
Facebook
iPhone
YouTube
eBay
Xbox 360
Ford
Yahoo!
Disney
Audi
432
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
Table 2
DISTRIBUTION OF TOTAL MENTIONS AND MENTIONS PER
BRAND, OFFLINE AND ONLINE
Category
Beauty products
Beverages
Cars
Children’s products
Clothing products
Department stores
Financial services
Food and dining
Health
Home design
Household products
Media and entertainment
Sports and hobbies
Technology
Telecommunications
Travel services
Percentage of Average Number of
Total Mentions Mentions per Brand
Number
of Brands Online Offline Online Offline
52
66
47
19
51
15
39
105
27
13
24
103
21
56
25
34
1%
3%
17%
0%
3%
4%
2%
4%
1%
1%
0%
32%
8%
17%
7%
1%
5%
13%
10%
2%
7%
5%
4%
12%
3%
2%
2%
9%
3%
12%
9%
3%
53,205
150,536
1,005,732
70,730
150,952
695,945
113,656
115,139
140,630
114,670
28,327
893,706
1,110,863
847,929
776,423
60,305
526
1,129
1,213
579
777
1,779
621
620
534
654
475
476
707
1,248
1,961
543
Notes: The sample contains only the most talked-about brands. The
online numbers contain mentions from all the available sources, whereas
the offline numbers only contain mentions from a weekly representative
sample of 700 people. As a result, the numbers for offline cannot be
directly compared with those for online.
people, whereas in the online data, we use a sample of posts.
This means that for the online data (like previous studies
that use online WOM data), we do not observe the receiving
side of the communication but rather only the “sender.” For
some purposes, this would mean a selection bias. For example, if we were measuring individual-level propensities to
engage in WOM, our sample has problems. However, for our
purposes—to measure aggregate brand mentions online—
this sample is appropriate. Another possible difference
between these two data sets is that different types of people
use the two channels. We acknowledge that differences in the
role of brand characteristics between the channels could be
due to these dissimilarities in people rather than in the channel, per se. After presenting our results, we discuss some ways
that future studies could leverage more refined measures to
provide a more disaggregate picture of WOM behaviors.
Brand Characteristics
To operationalize the brand characteristic variables identified in Figure 1, we used existing measurement scales
(e.g., Aaker’s brand personality) whenever possible. To collect the data, we conducted a large-scale original data collection using several existing public and proprietary databases. We then combined these sources (see Figure 2).
The first source is Y&R’s proprietary database, the Brand
Asset Valuator (YRBAV). It measures brand equity on four
perceived dimensions (termed “pillars” by the company):
energized differentiation, relevance, esteem, and knowledge. Y&R constructs this data set from a quarterly panel
survey that measures a broad array of perceptions and attitudes for a large number of brands, including 629 of the 697
brands we consider. From this survey, Y&R builds the four
pillars for each brand.
The second major source of data is from a survey we
developed and administered to a representative sample of
the U.S. population through Decipher Inc.5 We collected
data from 4,769 respondents on product involvement and
brand familiarity, excitement, complexity, visibility, and
perceived risk.
In addition, we used several other secondary sources.
First, we used Interbrand data on the brands ranked in their
“Top 100 Brands” list over the preceding few years. Second, we used the American Customer Satisfaction Index
(ACSI) to measure brand-level satisfaction. Third, we used
secondary data sources to code several other variables, such
as age and type of good. Next, we describe our variables,
scales, and measures in detail; their summary statistics
appear in Table 3.
Age. We define age as the time elapsed from the commercial launch of the brand to the reference current date, August
1, 2010. We obtained the data from brand publications and
from historical business and press data. Our oldest brand is
Colgate, launched in 1806, and the newest is the movie
Transformers: Revenge of the Fallen, released on June 29,
2009.
Type of good. We used Nelson’s (1974) and Laband’s
(1986) classifications to divide the brands into search,
experience, and credence goods. We operationalize this
measure, as originally defined, at a subcategory level
between the category and brand levels. For example, health
clubs and sports teams are subcategories within the category
of sports and hobbies. Using the definitions from the literature, two independent judges separately classified the subcategories. The intercoder agreement was 72%, and the
judges resolved all disagreements by consensus.
Complexity. We measured complexity in our survey using
a five-point scale based on Moore and Benbasat (1991) and
Speier and Venkatesh (2002). Our complexity scale includes
items regarding (1) the learning efforts needed to acclimate
to the brand, (2) the time required to fully understand its
advantages, (3) the difficulty of the product concept, and (4)
the mental effort required to use the brand (for our exact
questions, see Web Appendix B at www.marketingpower.
com/jmr_webappendix). In our brand list, Medicare is perceived as the most complex brand and Pledge as the least
complex.
Knowledge. We used two variables to measure the level of
knowledge about the brand. The first, familiarity, is a singleitem, five-point scale included in our survey that asks respondents to what extent they are familiar with the brand. The
second variable, knowledge, is one of YRBAV’s pillars. It is a
single-item, five-point scale that asks respondents to indicate
their level of intimate understanding of the brand. As Table 3
indicates, brands such as Band-Aid and Wal-Mart rank high
on familiarity and knowledge, whereas more local brands,
such as H-E-B Grocery and Shaw’s supermarket, rank low.
These two variables, though similar, differ in how detailed
or intimate the knowledge is. The correlation between these
5Decipher Inc. is a California-based company that specializes in developing and managing large-scale surveys. The questionnaire began with
screening questions about the respondent’s level of familiarity with the
category and the brands. Then, the system chose several brands with which
the respondent indicated familiarity and asked about the product and brand
attributes. The system dynamically allocated brands to respondents until
we reached 35–40 responses on each of our 697 brands. We describe an
annotated version of this complex questionnaire in Web Appendix B
(www.marketingpower.com/jmr_webappendix).
aThis
.40
.13
.33
.45
.17
.50
.16
.31
.36
.12
.67
.03
.43
.03
1.79
3.72
.76
.41
.25
.38
.59
.88
1.11
.21
.07
1.82
3.36
3.54
.16
.30
.5
.44
.72
.37
3.28
1.59
.49
.61
.51
.26
2.74
3.01
1.12
1.46
.43
.86
SD
0
.01
0
0
0
1.02
3.09
.04
0
0
1.01
1.48
.73
2.05
1.1
.17
.09
0
0
1.39
1.79
0
.12
Min
0
.29
0
0
0
1.54
3.52
.5
0
0
1.53
2.92
3.02
3.00
1.5
.39
.38
0
0
2.13
2.78
.03
.24
25%
0
.58
0
0
0
1.81
3.62
.96
0
0
1.81
3.42
3.71
3.27
1.63
.46
.55
1.00
0
2.65
3.02
.08
.41
50%
0
1.00
0
1.00
0
2.01
3.97
1.61
0
0
2.06
3.79
4.18
3.54
1.69
.57
.77
1.00
1.00
3.24
3.27
.33
.84
75%
is the average of the two excitement items from the Aaker questionnaire as described in the “Data” section.
Dependent
Online brand mentions (/1,000,000)
Offline brand mentions (/1,000)
Social
Differentiation
Esteem
Middle (premium/value)
Premium (premium/value)
Relevance
Visibility
Emotional
Excitementa
Satisfaction (/50)
Functional
Age (/50)
Search
Credence
Complexity
Familiarity
Knowledge
Hybrid
Perceived risk
Involvement
Controls
Interbrand top 100
Usage (/50)
Mixed (product/service)
Service (product/service)
Internet brand
M
Table 3
SUMMARY STATISTICS
1.00
1.79
1.00
1.00
1.00
2.62
4.38
4.09
1
1
3.03
4.62
5.16
4.51
1.79
1.12
1.67
1.00
1.00
4.75
3.99
14.58
15.04
Max
Dr. Pepper
Beverages
Porsche
Band-Aid
Pledge
H-E-B Grocery
Shaw’s
613
613
613
613
613
613
613
613
613
613
613
613
613
613
201
613
613
613
613
613
613
613
613
Observations
Used
Transformers: Revenge of the Fallen
Medicare
Charter Communications
Saab
Lamborghini
Days Inn
Ugly Betty
Vault Energy Drink
America’s Got Talent
Minimum
Value Brand
Medicare
Financial services
Medicare
Band-Aid
Wal-Mart
Colgate
iPhone
Heinz
Kraft
Microsoft
Food Network
Tylenol
Google
Coca-Cola
Maximum
Value Brand
On Brands and Word of Mouth
433
434
variables is .80. Therefore, we use principal component
analysis to identify a single factor to incorporate both these
variables. This one factor explains 91% of the variation, and
both variables load positively (for details, see Web Appendix C at www.marketingpower.com/jmr_webappendix).
Differentiation. To measure differentiation, we used the
YRBAV pillar energized differentiation. Energized differentiation is a weighted average of items indicating to what
extent the product is different, distinctive, unique, dynamic,
and innovative, a fairly direct of measure of differentiation.
Of our list of brands, Food Network has the highest differentiation score, and Days Inn has the lowest.
Relevance. We measure how relevant a brand is to a
broad set of people with the YRBAV pillar relevance. This
pillar measures the percentage of people who stated that the
brand is personally appropriate for them. Kraft is the most
relevant on our list, and the car brand Saab is the least.
Quality. We measure quality through the final YRBAV
pillar, esteem. This variable captures the extent to which
people hold a brand in high esteem. We measured this
variable through items asking about the leadership, reliability, and quality of the brand. Tylenol has the highest esteem
score and the prime-time television soap opera Ugly Betty
has the lowest.
Premium. Two independent judges classified each brand
as one of the following: premium, value, or middle. The
intercoder agreement was 70%, and the judges resolved all
disagreements by consensus. They classified brands relative
to the product type (e.g., they evaluated Clinique relative to
beauty products and Hilton with respect to other hotels). In
formulating these classifications, the judges used secondary
data on various aspects such as the relative price to the
category.
Visibility. We measure visibility as Rogers’s (1995)
observability construct, using a five-item, five-point scale
based on Moore and Benbasat (1991). These survey items
determine whether respondents commonly see the brand in
their environment. The brand with the highest visibility on
our list is Microsoft, and Lamborghini received the lowest
visibility score.
Excitement. We included in our survey a subset of
Aaker’s (1997) five-point excitement scale, which includes
items such as “exciting” and “spirited.” The full scale comprises items that overlap with other variables in our analysis
(e.g., age, differentiation) and, as a result, leads to inflated
standard errors (i.e., multicollinearity). Note that our qualitative results do not change if we use the full excitement
scale. As Table 3 indicates, the most exciting brand on our
list is iPhone and the least exciting is Medicare.
Satisfaction. We use the ACSI, a standard measure of satisfaction for American corporate brands (Fornell et al.
1996). The measure is a 0–100 index collected each quarter
using 250 customer telephone interviews per brand on a
rolling set of brands, with each receiving at least one measure each year. Of our list of brands, 209 have an ACSI score
(with Heinz having the highest score and Charter Communications the lowest). Subsequently, we discuss how we
handled this missing data challenge.
Perceived risk. Rogers (1995) defines perceived risk as
the functional, financial, and emotional uncertainty associated with the product (in which emotional uncertainty is the
feeling of social embarrassment that might be associated
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
with using the brand). We use the full three-item, five-points
scale (Ostlund 1974) and collect this measure of perceived
risk in our survey. Of our list of brands, Medicare has the
highest perceived risk score, and Dr. Pepper is perceived as
the least risky.
Involvement. To measure involvement, we use Ratchford’s
(1987) three-item, five-point scale. The items measure the
importance of the purchase decision, the amount of thought
invested in the decision, and the consequences of making
the wrong decision. In line with prior studies, our measure of
involvement (collected by survey) is at the category level. In
a preliminary check, we measured involvement at the brand
level but observed little variation between brands within a
category. Of our 16 categories, financial services have the
highest involvement level and beverages have the lowest.
Control Variables
We also include control variables to account for a variety
of other concerns. For example, one might argue that people
talk about some brands simply because these brands are
widely used or have existing brand equity (e.g., high media
coverage or ad budgets).
Brand equity. We use data from Interbrand to measure
brand equity and to capture advertising and media coverage
effects. Using Interbrand’s list of top 100 brands from 2008–
2010, we code a binary variable indicating whether the brand
is on the list. We expect brand equity to increase WOM.
Usage. To gauge usage, we use a measure from YRBAV’s
survey of the percentage of people who answered that they
use the brand frequently or occasionally. Band-Aid is scored
the highest on usage and Porsche is scored the lowest.
Product/service. Two independent judges classified each
brand on the list to one of the following: product, service, or
mixed. The judges used the four criteria of Parasuraman,
Zeithaml, and Berry (1985): intangibility, inseparability (of
production and consumption), perishability (cannot be
inventoried), and heterogeneity (difficult to standardize).
Accordingly, the judges classified video games and movies
as products, fashion brands that are sold both in their own
stores and in other outlets as mixed, and sports teams as
services. The intercoder agreement was 82%, and the judges
resolved all disagreements by consensus.
Internet brand. Seventeen of the brands on our list, including eBay, Amazon, Expedia, and Google, are Internet-based
services and, thus, by their nature might be more relevant for
online than WOM. To control for this factor, we code a binary
variable indicating whether the brand is an Internet brand.
Final Sample and Data Summary
Some of the brands included in our initial list (i.e., the top
700 talkable brands) are not included in the final sample
because many of our variables are not available for them.
For example, movies and television programs and some
subbrands (e.g., Cherry Coke) do not have data available for
satisfaction or any of the YRBAV variables. As a result, our
final sample contains 613 brands. For these brands, we have
complete data on all variables but satisfaction (which is
available for only 209 brands). This final data set contains
two dependent variables, online and offline brand mentions
(WOM), and 19 explanatory variables. Table 3 displays
summary statistics for the dependent and explanatory
variables, and Table 4 presents the correlations for the
Differentiation
Esteem
Middle (premium/value)
Premium (premium/value)
Relevance
Visibility
Excitement
Satisfaction
Age (/50)
Search
Credence
Complexity
Knowledge factor
Perceived risk
Interbrand 100
Usage (/50)
Mixed (product/service)
Service (product/service)
Internet brand
.10
–.15
.30
–.01
.09
.57
.19
–.31
.10
–.15
.12
0
.07
.22
–.03
.17
–.09
.11
1
.10
1
.05
–.05
.80
.51
–.15
.23
.35
.04
–.03
–.42
.68
–.49
.22
.58
–.03
–.25
–.01
–.15
.05
1
–.58
.04
.09
–.15
–.16
.07
.02
.17
–.02
.02
–.06
.01
0
.03
–.12
.08
.3
–.05
–.58
1
–.17
–.10
.26
.17
–.03
0
–.09
.17
–.11
.13
.08
–.16
.01
.04
–.05
–.01
.80
.04
–.17
1
.54
–.20
.25
.25
.03
–.11
–.62
.70
–.67
.08
.85
–.07
–.18
0
.09
.51
.09
–.10
.54
1
.07
.13
.11
.07
–.15
–.50
.51
–.50
.17
.45
.01
–.20
–.06
Middle Premium
Different(Premium/ (Premium/
iation
Esteem
Value)
Value) Relevance Visibility
Table 4
.57
–.15
–.15
.26
–.20
.07
1
.24
–.29
.10
–.20
.08
–.10
.02
.07
–.13
.09
–.07
–.01
.19
.23
–.16
.17
.25
.13
.24
1
.13
.07
–.10
–.54
.23
–.60
.06
.23
.03
–.61
–.18
Excitement Satisfaction
–.31
.35
.07
–.03
.25
.11
–.29
.13
1
–.01
.09
–.16
.24
–.18
.11
.14
–.01
–.12
–.16
Age
(/50)
CORRELATIONS
.10
.04
.02
0
.03
.07
.10
.07
–.01
1
–.13
.02
–.03
.04
.08
.01
.13
–.15
–.06
Search
–.15
–.03
.17
–.09
–.11
–.15
–.20
–.10
.09
–.13
1
.31
–.17
.17
–.02
–.18
–.05
.12
.07
.12
–.42
–.02
.17
–.62
–.50
.08
–.54
–.16
.02
.31
1
–.56
.81
.04
–.65
.01
.35
.14
.68
.02
–.11
.70
.51
–.10
.23
.24
–.03
–.17
–.56
1
–.52
.13
.72
–.04
–.17
–.01
0
.07
–.49
–.06
.13
–.67
–.50
.02
–.60
–.18
.04
.17
.81
–.52
1
0
–.64
.07
.36
.06
Knowledge Perceived
Credence Complexity Factor
Risk
.22
.22
.01
.08
.08
.17
.07
.06
.11
.08
–.02
.04
.13
0
1
–.01
.04
–.16
.09
–.03
.58
0
–.16
.85
.45
–.13
.23
.14
.01
–.18
–.65
.72
–.64
–.01
1
–.07
–.08
.01
Interbrand
Top
Usage
100
(/50)
.17
–.03
.03
.01
–.07
.01
.09
.03
–.01
.13
–.05
.01
–.04
.07
.04
–.07
1
–.16
–.03
–.09
–.25
–.12
.04
–.18
–.2
–.07
–.61
–.12
–.15
.12
.35
–.17
.36
–.16
–.08
–.16
1
.18
.11
–.01
.08
–.05
0
–.06
–.01
–.18
–.16
–.06
.07
.14
–.01
.06
.09
.01
–.03
.18
1
Mixed
Service
(Product/ (Product/ Internet
Service) Service) Brand
On Brands and Word of Mouth
435
436
explanatory variables. These correlations use the full set of
brands in our analysis, with the exception of satisfaction,
which we calculated using only the 209 brands for which
satisfaction is observed.
Our data are aggregate and from multiple sources. This
means that we do not observe how the brand perceptions of
a specific person are translated into his or her specific
WOM. However, these multiple sources also mean that different sets of people answered our variables. This separation implies that our analyses are protected from common
method variance. In particular, false correlations due to a
single measurement system or sampling variation cannot
explain our results.
ESTIMATION AND RESULTS
This section describes the empirical model and the
estimation results. We also include an analysis of the role
of the brand characteristics, the overall importance of the
three drivers, a content analysis, and numerous robustness
checks.
Empirical Model and Estimation Procedures
The formal model describes a set of brands i = 1, 2, ..., N,
each belonging to one of K categories indexed by k. The
dependent variables are counts of brand mentions. Counts
are typically treated as having a nonnormal distribution; following this practice, we use a negative binomial distribution to model the mentions. Specifically, the probability
density of WOM brand mentions from channel m for brand
i in category k is expressed by the following equation:
m
m
m
ym
ik ~ fNegBin(g k + b Xik, a ),
where fNegBin is the density of the negative binomial with
dispersion parameter am, which varies by online and offline
channels and mean parameter g mk + bmXik. The mean
parameter incorporates (1) the vector Xik that includes the
variables of interest and controls, (2) the channel-specific
linear parameters bm, and (3) the channel-specific category
level effects g m
k.
One variable, satisfaction, has a large number of missing
values. The reasons are unrelated to the variable’s role in
WOM, but dropping all observations with missing values
would reduce our sample size too severely (by two-thirds).
Therefore, we assume a prior for the missing data and use a
missing at random assumption to impute values for the
missing observations. Specifically, we denote by I the set of
observations that are incomplete (i.e., missing values for
satisfaction) and by C the set of observations that are complete and let the prior of i ΠI follow a normal distribution
parameterized by the first two moments of the complete data:
X1ik ~ fN[XC, V(XC)],
where the function fN is the normal density, X1ik are the
incomplete observations of satisfaction, XC is the mean of
the complete data, and V(XC) is the variance of the complete
data. Note that whereas the prior is only based on the complete satisfaction data, the posterior distribution is influenced by the full model likelihood. As a result, and because
the observations in I are incomplete only with respect to one
variable, the posterior distribution of the imputed data, X1ik,
also depends on the relationship to all the other variables.
We note that this Bayesian approach is consistent with the
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
likelihood-based approach Schafer and Graham (2002)
suggest and naturally accommodates multiple imputations
through the posterior simulation. Web Appendix D (Part
D.3; www.marketingpower.com/jmr_webappendix) presents
robustness checks against alternative imputation procedures.
To complete the model, we describe the other priors,
beginning with the category-level effect. Our brand observations originate from a variety of categories. Different
categories may generate more or less WOM on average.
Some of this heterogeneity might be explained by the (only)
category level variable in the analysis, involvement. The
remaining heterogeneity is random from our perspective.
Thus, we use a multilevel model, enabling the category
level effects to be a function of involvement, an overall
average, and a random effect. Specifically, the prior distribution for the kth category-level effect on channel m WOM,
gm
k is
2
m
gm
k ~ fN(d Zk, s m),
where dm is a row-vector of parameters, s2m is the variance
parameter, and the vector Zk includes an intercept and the
involvement variable. We place priors on the parameters qm =
{bm, dm, am, s2m} as follows:
bm ~ fN(bm, A–1); am ~ fGAM(a0, b0);
–2
dm|s 2m ~ fN(dm, s 2mA–1
d ); s m ~ fc2(h0, v0).
The distribution fN is the multivariate normal distribution of
same dimension as the mean vector and fGAM is the gamma
distribution. We refer to this joint prior on the parameters
qm as (qm) and note that we use standard values for the
prior arguments to generate diffuse priors.
Thus, the complete posterior likelihood, Lm, is proportional to
 n
m
m

fNegBin γ m
k + β X ik , α

 i=1
∏
(

×


K
)  ∏ fN  XC , V( XC ) 


  i ∈ I


2 
π θm .
fN δ m Zk , σ m

k =1

∏ (
) ( )
We estimate the model using Markov chain Monte Carlo
posterior simulation. Web Appendix E (www.marketingpower.com/jmr_webappendix) presents the details related
to the estimation.
Results from the Full Model
The full model results appear in Table 5. Next, we organize our discussion of the results by the drivers.
The social driver. We begin our discussion with our focal
variable related to the social driver, the level of product differentiation. Recall that our theoretical framework identified this characteristic (in relation to the desire to express
uniqueness) and that it was not previously studied in the
context of WOM. As the theoretical framework predicted,
differentiation (measured by YRBAV’s energized differentiation pillar) has a positive and significant effect on WOM
both on- and offline. This means that people tend to talk
more about differentiated brands than other brands. Notably,
as we suggested in the “Theoretical Framework” section,
the effect is much stronger in the online setting than in
On Brands and Word of Mouth
437
Table 5
ESTIMATION RESULTS
Variable
Social
Differentiation
Esteem
Middle (premium/value)
Premium (premium/value)
Relevance
Visibility
Emotional
Excitement
Satisfaction
Satisfaction^2
Functional
Age
Search
Credence
Complexity
Knowledge factor
Hybrid
Perceived risk
Involvement
Controls
Category average
Interbrand top 100
Usage
Mixed (product/service)
Service (product/service)
Internet brand
Dispersion
Posterior Mean
1.78**
1.22**
.50**
.47**
–.26
.92**
.71**
4.60*
–3.56**
.13
–.30**
–.01
–.49*
.49*
.91**
–.58
8.36**
.95**
–1.07**
–.36
.54**
.31
3.12**
Online
(95% Credible Interval)
(.90, 2.65)
(.66, 1.79)
(.31,
.69)
(.19,
.75)
(–.62,
.06)
(.65, 1.17)
(.39,
.99)
(–.54, 9.75)
(–5.25, –1.94)
(–.07,
.37)
(–.56, –.06)
(–.37,
.37)
(–.98,
.05)
(.33,
.65)
(.30,
(–2.01,
(1.67,
(.74,
(–2.22,
(–.99,
(.23,
(–.08,
(2.75,
15.37)
1.17)
–.13)
.34)
.79)
.70)
3.51)
*Significant at the 5% level (i.e., 95% credible interval does not overlap 0).
**Significant at the 10% level (i.e., 90% credible interval does not overlap 0).
offline conversations (1.78 vs. .62). Recall that this might
be due to two fundamental differences between online and
offline environments. First, in offline interactions, a person
has many ways to communicate uniqueness (e.g., by wearing the branded clothes), and brand name-dropping is less
necessary for this purpose. Second, online interactions
involve broadcasting to a wider audience than most offline
communications. As a result, online WOM may involve
communicating with many people who are less personally
familiar with the communicator, leading to a stronger desire
to express personality and especially uniqueness. Differentiation is not the only characteristic for which significant
differences exist between online and offline settings. We
provide a comprehensive picture of these differences in the
“Online Versus Offline” subsection to follow.
The second motive under the social driver is the desire to
enhance oneself by associating with high-quality products
to demonstrate expertise (measured by esteem) and signal
higher status (through premium products). The results for
esteem are consistent with these expectations for both
online and offline settings, indicating that brands with
higher perceived quality are mentioned more often. For premium products, the effect is only significant online. Specifically, we find that relative to value brands, people talk more
online about premium and middle-premium brands.
The final social motive, the desire to converse, is measured by visibility and relevance. For visibility, we find the
expected positive effect for both channels, indicating that
1.44)
.85)
Posterior Mean
.62**
.52**
.01
–.07
.29**
.72*
Offline
(95% Credible Interval)
(.16, 1.12)
(.22, .81)
(–.09, .10)
(–.21, .06)
(.10, .47)
(.53, .91)
.44**
5.59**
–3.10**
(.27, .60)
(3.21, 8.17)
(–3.93, –2.30)
.03
.13
(–.26, .32)
(–1.01, 1.25)
–.17**
.04
–.60**
.43**
.46**
–.73
.26**
–.84**
.27
.62**
–.30**
8.36**
(–.29, –.05)
(–.11, .27)
(–.81, –.39)
(.09, .76)
(.36, .56)
(–5.34,
(.14,
(–1.36,
(–.15,
(.49,
(–.49,
(7.44,
4.01)
.39)
–.28)
.62)
.77)
–.10)
9.34)
more visible brands are mentioned more often. This result is
consistent with Berger and Schwartz (2011) and generalizes
their finding to a larger set of brands and categories as well
as for both on- and offline channels. For relevance, we find
a significant, positive effect offline but an insignificant,
negative effect online (for further discussion of such differences, see the subsection “Online Versus Offline”).
The emotional driver. The emotional driver includes two
characteristics, excitement and satisfaction. We first consider our focal variable for this driver, excitement, which
previous research has not explored in the context of WOM.
As expected, we find that more exciting brands receive
more WOM and that the effect is strongly significant for
both online and offline channels. We interpret this result to
mean that when consumers are excited about a brand, they
are likely to experience emotional arousal that leads them to
speak with others.
The role of satisfaction is more complicated. In line with
prior research (Anderson 1998; Richins 1983), we expected
that at extremely low and high levels of satisfaction, consumers are much more likely to mention brands, leading to
a U-shaped effect. To capture this potential shape, we
included in the model linear and quadratic terms for satisfaction. However, in both the on- and offline channels, we
find a monotonic concave effect. For the observed values of
satisfaction (between 55 and 89), we find that as satisfaction increases, WOM decreases. This result means that the
data support the greater WOM at low satisfaction levels but
438
do not support greater WOM for high satisfaction levels. It
is possible that previous findings about the high WOM levels at high satisfaction levels were due to the exclusion of
variables that are related to satisfaction, such as esteem and
excitement, which we included in our model. In other
words, our analysis studies the role of satisfaction beyond
the effect of these variables.
The functional driver. The functional driver relates to the
need to obtain and the tendency to provide information. We
begin by discussing our focal variable for this driver, complexity, which has not been studied in the context of WOM.
The effect of complexity in the offline channel has the
expected positive sign and the estimates are statistically significant, but in the online channel, the effect is negative and
marginally significant (p < .1). In other words, people talk
more in the offline world about brands that are more complex, but in an online environment, they talk more about
brands that are less complex. Notably, we find a similar pattern (significant and negative offline; insignificant online)
for age, indicating that people are more likely to discuss
newer brands than older brands offline. Finally, for type of
good, we find that for online channels, search goods are
mentioned statistically less often than experience goods; yet
for offline channels, credence goods are mentioned statistically less often than experience goods.
As for the information supply variables, we find, as
expected, significant positive effects for the knowledge factor, meaning that people share more information about
brands they are familiar with and knowledgeable about. This
tendency is qualitatively the same across the two channels.
We also conducted an analysis in which the coefficients
of the model are allowed to differ across the types of good
(search, experience, and credence). This moderation analysis, presented in Web Appendix F (www.marketingpower.
com/jmr_webappendix), illustrates that our results are relatively robust to such an extension, but at the same time, it
also suggests that evaluating such moderation effects could
be a fruitful line of research.
The hybrid characteristics. As we discussed previously,
two characteristics (perceived risk and involvement) do not
fit well into a single driver and thus are considered “hybrid”
(i.e., they contain elements of multiple drivers). As we
expected, the effect of perceived risk is positive. It is highly
significant in the online model but not significant in the
offline model. We expected involvement to have a positive
effect, but because we measured it at the category level and
with only 16 categories, the limited variation did not allow
us to estimate the effect adequately. We do not find a significant effect in any of the models.
Controls and dispersion. All our control variables are
highly significant; brands in the Interbrand top 100 have
higher WOM, brands with higher usage have less WOM,
services get more WOM than products, and Internet brands
receive less WOM offline. Finally, the dispersion parameter
is higher in the offline than the online channel, reflecting the
larger dispersion in the number of online mentions. This is a
characteristic of the measurement system and modeling
approach and not reflective of any actual differences across
the two channels.
Online versus offline. Although some variables have a
similar coefficient in the online and offline settings (e.g., the
coefficients of the knowledge factor are .49 and .46, respec-
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
tively), others differ meaningfully either in their coefficients
or in their significance between the two settings (or both).
To present a clear picture of these differences, we discuss
them together here. Note that for each variable, we can
directly compare the coefficients from the on- and offline
regressions because the dependent variables are logged and
the independent ones are the same. As a result, in both
regressions, the coefficients represent the percentage
change in WOM for a unit change in the variable.
We begin with the social driver, for which we find the
most dramatic differences. We previously pointed out that
differentiation has a much stronger effect online than offline
and discussed some possible explanations for these differences. We find a similar difference for esteem, for which the
coefficient is significant both online and offline but is more
than twice as large online. Furthermore, there is a similar
finding with respect to premium. The results for both esteem
and premium are consistent with the idea, discussed previously, that people seek to enhance themselves more often
online than offline, perhaps due to the lack of nonverbal
cues (e.g., clothing brands worn) to help signal status and
identity as well as the broadcasting nature of the online
medium. In contrast, visibility is similarly strong both online
and offline, whereas relevance is significant and positive
offline but is negative and not significant online. This difference could be due to the diversity of tastes of the broader
online audience, which implies that even brands with lower
relevance are still relevant for many people and thus can serve
as conversation material; for example, a fan of an indie band
may find few fans offline but many online, and thus, the
band’s low relevance does not suppress online WOM.
The coefficients of the emotional driver are essentially
the same across channels, so we next turn to those of the
functional driver. Here, the most noteworthy variables are
age and complexity. Less complex brands have more WOM
online, whereas more complex brands get more WOM
offline. A possible explanation for this difference hinges on
the advantages of offline conversations in clarifying complex issues, because such conversations are truly interactive
and enable quick responses and clarifications. In contrast,
online conversations are more likely to be asynchronous
(Morris and Ogan 1996) and take more time to respond,
clarify, and exchange information. As a result, exploring
new or complex features of a brand may be easier offline
than online. This same argument can explain the similar
result for the age variable.
That said, as we illustrate subsequently (see the “Robustness Checks” subsection), our results are robust, but the
estimate of complexity is perhaps least so. Furthermore,
some might argue that the lack of effect online (and even its
marginally significant opposite effect) is because we did not
observe the people who passively browse (i.e., receive but
do not post) online (for the differences between WOM generation and consumption, see Yang et al. 2012). By considering both online and offline data together, we can empirically
observe the potential effects of this possible shortcoming of
typical online data sources.
Results on the Relative Importance of the Three Drivers of
WOM
To compare the importance of the variables under the
three drivers at the same time, we determine what happens
On Brands and Word of Mouth
to the fit of the model when we exclude each of these drivers
from the analysis. In other words, we examine models with
subsets of the variables corresponding to all combinations
of the drivers. To compare these submodels, we present the
model log marginal likelihoods (LML).
Before proceeding to the results, we highlight two points
about this exercise. First, satisfaction requires a missing
data model, which induces much greater variation in the
LML and, as a result, does not allow us to compare across
subsets of drivers. Therefore, we exclude it from this analysis. This exclusion could lead the importance of the emotional driver to be understated. Second, the hybrid characteristics could belong to both the functional and the
emotional drivers. Thus, we use submodels with and without the hybrid characteristics to examine the overall role of
the three fundamental drivers.
Table 6 presents the results of this analysis (see also Web
Appendix G at www.marketingpower.com/jmr_webappendix).
The most notable finding here is the difference between
online and offline channels. We find that for the online
model, the order of importance of the drivers is social, functional, and emotional. Overall, the importance of the social
and functional drivers is significantly greater than that of
the emotional driver. For the offline model, the order is
emotional, functional, and social; the importance of the
emotional driver is significantly greater than the other two
drivers. In other words, whereas the emotional driver is the
most important in offline conversations, the social driver is
the major force in online brand mentions. These results portray a nuanced and insightful picture of WOM. One interpretation argues that offline conversations, which typically occur
in one-on-one settings, are more personal and intimate and
thus allow people to share emotions such as excitement and
satisfaction. In contrast, nonverbal signals are not available
for online WOM, which usually involves broadcasting to
Table 6
RELATIVE IMPORTANCE OF THE FUNCTIONAL, SOCIAL, AND
EMOTIONAL DRIVERS
Models with the Hybrid Characteristics
Social
Emotional
Functional
Social and emotional
Functional and emotional
Functional and social
Models Without the Hybrid Characteristics
Social
Emotional
Functional
Social and emotional
Functional and emotional
Functional and social
Online LMLa Offline LML
–8,393.6
–8,455.9
–8,439.4
–8,376.5
–8,424.1
–8,392.4
–8,381.8
–8,453.5
–8,454.2
–8,387.3
–8,412.6
–8,358.4
–5,745.4
–5,312.0
–5,374.4
–5,640.2
–5,536.9
–5,899.7
–5,707.7
–5,316.1
–5,408.8
–5,573.1
–5,363.8
–5,867.0
aLML with higher (less negative) values, indicating better fit to the data.
Notes: This table indicates that for submodels that contain only one
driver, in the online environment, the social driver fits best (LML = –8,394)
and functional fits second best (LML = –8,439), whereas in the offline
environment, the emotional driver fits best (LML = –5,312) and functional
fits second best (LML = –5,374). The same relationship holds for submodels that contain one driver and including the hybrid motives. For submodels containing two drivers, the best models online contain social motives
and the best models offline contain emotional motives. This pattern is true
for both models with and without the hybrid characteristics.
439
many people (e.g., Twitter) and may be more appropriate
for social signaling (e.g., communicating uniqueness).
An alternative explanation of these results is that they are
driven primarily by the difference in measurement between
online and offline channels. In particular, recall effects may
exist in the offline data but not the online data, and it is possible that such effects may lead to greater recall for brands
with high levels of emotional characteristics. Although we
cannot rule this alternative out, some aspects of the data
minimize this possibility. First, the self-reports document
only one day, and respondents are requested to keep the
diary with them at all times so that we could minimize the
lag between occurrence and reporting. Second, the data
provider (the Keller Fay Group) ran additional checks in
which respondents used smartphones and recording devices
to examine whether pure observational behavior differed
from the self-reports. They found that the pattern of
observed WOM was similar to the self-reported one.
Results on the Connection Between Brand Characteristics
and WOM Content
Although our study relates brand characteristics with
WOM mentions, it is possible that these characteristics also
relate to WOM content. For example, it is possible that
exciting brands, such as Arizona Beverage Company, not
only receive more WOM but also receive WOM that
expresses more excitement about the brand. Although
extending our theoretical and empirical framework to
address such an issue is beyond the scope of this study, it is
worthwhile to observe whether such a content–characteristic
connection is likely, using a small-scale test.
For such a test, we focused on (1) three characteristics
(excitement, esteem, and differentiation), (2) ten major
product categories, and (3) 41 brands that we selected to
provide a range of scores above and below average for these
three characteristics. Using a commercial text-mining tool
(NetBase Solutions’s Insight Workbench tool) on online
data, an independent coder identified for each category and
characteristic a set of words and phrases (called “themes”)
that could describe this characteristic in the context of the
category (e.g., for excitement and cars: “great experience,”
“popular model,” and “head turning”; for excitement and
beauty products: “newest beauty obsession”). We then used
the text-mining engine to count how often each theme
appeared in the brand’s WOM mentions as well as the total
brand mentions over 365 days from March 2012 to March
2013. Thus, we had a “content score” for each brand and
characteristic, calculated by the number of times respondents mentioned this characteristic with respect to this
brand, divided by the total mentions.
Using these data, we ran a brand-level (subscripted by i)
logistic regression with fixed effects for each of the ten categories (subscripted by k). We estimated the model jointly for
all content scores but only allowed the relevant characteristic to affect the content score (e.g., excitement affects the
excitement content score). Thus, the model for excitement is
Logistic(Excitement Content Scorei(k)) =
b0k + b1Excitementi(k) + ei(k).
Table 7 presents the results. The coefficients for all three
characteristics are positive, and two of three are significant,
440
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
Table 7
THE RELATIONSHIPS BETWEEN BRAND CHARACTERISTICS
AND CONTENT SCORE
Content Score
Log (excitement)
Log (esteem)
Log (differentiation)
*p < .05.
Excitement
.29* (.12)
Brand Characteristic
Esteem
.01 (.13)
Differentiation
.43* (.21)
suggesting that brand characteristics increase the proportion
of brand mentions that involve content that is related to the
specific brand characteristic. Although these results are preliminary, they provide encouraging initial support that brand
characteristics might have a role that extends beyond a general increase in WOM.
Robustness Checks
In addition to the preceding analyses, we also conducted
a range of robustness tests to ensure that our results are not
influenced severely by selection biases, multicollinearity,
outliers, or the missing data model. We discuss the complete
set of analyses in Web Appendix D (www.marketingpower.
com/jmr_webappendix) and note a few highlights here.
Selection bias. Our analysis is based on the 600+ most
talkable brands. How sensitive are the results to this selection? We can get a sense of the selection issue by decreasing the number of brands to the top 550, 500, 450, and 200.
We find that in all but the smallest data set, the order of the
three drivers remains the same; that is, the social driver is
most important for online WOM, and the emotional driver
is most important for offline WOM. Some coefficients
change in effect and significance as we decrease the sample
size, but overall, the significant effects are relatively robust
to sample selection biases. See Web Appendix D, Part D.1
(www.marketingpower.com/jmr_webappendix) for details,
including the richness of results across the different sets of
brands. For example, age might play a more important role
and visibility a less important role in the top 200 brands
than in the others.
Multicollinearity. Our analysis of multicollinearity indicates that it has no meaningful impact on our results with
respect to either the relative importance of the three drivers
or the specific coefficient estimates. See Web Appendix D,
Part D.2 (www.marketingpower.com/jmr_webappendix),
for details.
Missing data. First, we note that our approach to modeling the missing data is consistent with the recommendation
for such situations (Schafer and Graham 2002). Specifically,
we use a Bayesian (likelihood-based) approach that imputes
the missing data by sampling from the posterior distribution
that depends on all the other data. Second, we note that our
analysis of the relative importance of the three drivers does
not employ the missing data model. Third, to obtain yet
another view of robustness, we apply two alternative
approaches: case deletion and single conditional imputation.
Case deletion uses far fewer observations, and as a result,
fewer variables are significant; however, given the smaller
sample size, the results are remarkably similar. Conditional
imputation ignores the errors in the missing data model but
comes close to our full model results. Overall, these robustness checks for the missing data model suggest that our
results are not driven primarily by this approach (for details,
see Web Appendix D, Part D.3, at www.marketingpower.
com/jmr_webappendix).
Outliers. We find relatively minor outlying cases (with
absolute standardized residuals < 4). Dropping these cases
had no impact on the statistical significance or direction of
effects (for details, see Web Appendix D, Part D.4, at www.
marketingpower.com/jmr_webappendix).
DISCUSSION
Although brands and WOM are two fundamental marketing concepts, prior research has largely ignored their relationship. Here, we show that they are closely related and
demonstrate that brand characteristics play an important role
in explaining the level of WOM. Furthermore, these results
are consistent with the theoretical framework we present,
which posits that the brand characteristics affect WOM
through three drivers: social, emotional, and functional.
The results portray a nuanced, intricate picture of the
brand–WOM relationship in two aspects. First, all three
drivers—social, emotional, and functional—play a role in
this relationship. In other words, WOM is not related to
only one characteristic (e.g., perceived risk, visibility) or
driver. All the brand’s facets are involved. Second, the role
of brand characteristics differs across the WOM channels.
For example, new brands are more talked about offline, but
we find no support for this relationship online. In contrast,
premium brands have significantly more online brand mentions, whereas we find no support for such a relationship
offline. Furthermore, the channels differ in what fundamental drivers are most important to WOM: whereas the order
of importance of the three drivers in the online channel is
social, functional, and emotional, the order for the offline
channel is emotional, functional, and social.
Managerial Implications
Until the early 2000s, WOM was largely considered a
side effect of marketing activity. Today, marketers are trying to develop a systematic approach to manage it.6 Our
work can assist in this task. As we demonstrate in the following four points, this research can provide practitioners
with tools in planning, measuring, and managing not only
their WOM initiatives but also their branding practices and
marketing mix as a whole.
Connecting brand characteristics and WOM. Marketers
are interested in creating talkable brands (“Creating Talkable Brands” was the title of WOMMA’s annual summit in
2012). Our findings can assist them in identifying the brand
characteristics that can accomplish this. A brand manager
who wants high WOM can now evaluate which characteristics to use to design WOM into a brand. Consider the case
of visibility. A firm that developed a new type of digital
music player for cars may have a technological option to
embed this player deep in the dashboard or make it a more
visible component of the interior. Because visibility
enhances WOM and because our model can project the
6In this context, for example, practitioners debate whether WOM and
advertising are complementary to or substitute for each other (for a review,
see Armelini and Villanueva 2010).
On Brands and Word of Mouth
magnitude of the effect, a brand manager may be able to
weigh the total costs and benefits of the design choice. The
“Intel Inside” campaign of 1999 did something similar: it
increased the visibility of the microprocessor and contributed to the firm’s WOM (Intel is on our list of 700
brands).
To assist managers in such an evaluation task, we created
a model-based Descriptive Decision Support System
(DDSS) in Microsoft Excel (Power and Sharda 2009). To
use it, managers would need to conduct a survey among
consumers in their target market to measure the brand characteristics and enter these into the DDSS to obtain the
expected level of WOM.
This method can be also used as a diagnostic tool for one
dimension of brand health, a diagnosis that is of increasing
importance to brand managers (Berg, Matthews, and
O’Hare 2007). Specifically, by comparing the expected
level of WOM to the actual level, it is possible to determine
whether the actual level is above or below the expected
WOM and test whether the brand lives up to its WOM
potential. As Figure 3 indicates, on the one hand, Dove,
T.G.I. Friday’s, and Coca-Cola do exceptionally well offline
(i.e., there is a large positive gap between actual and
expected WOM). On the other hand, AOL, Charter Communications, and Mug Root Beer do poorly both offline and
online compared with what we would expect on the basis of
their brand characteristics. Notably, brands such as Facebook,
Staples, and Cheerios meet or exceed the expectations online
but underperform offline. Managerially, this could call for
more efforts to exploit these brands’ WOM potential in the
offline environment (which is still the highest-volume WOM
channel). Note that the focus here is on the performance
relative to expectations rather than on absolute values. Low
levels of WOM might not necessarily indicate a problem;
some brands, given their characteristics, cannot expect high
levels of WOM. This is especially evident in the case of
categories such as financial products. Indeed, awareness of
Figure 3
ACTUAL VERSUS PREDICTED PERFORMANCE FOR TOP 2%
OF OVERPERFORMING AND UNDERPERFORMING BRANDS
(7,01,02304*+5#()*+,#
(7
7,01,02304*+5#()*+,#
.+/,01,02304*+5#(+-*+,
.+
+/,01,02304*+5#(+-*+,
'#
!'#
"##
!"#
()*+,#
()*
+,#
(7,01,02304*+5#
,02304*+5#
(7,01
(+-*+,#6+/#()*+,
(+
-*+,#6+/#()*+,
<1@&'
<1@
&'
9$:&B$%;'
9$:&B$%;' 6
678'9#$:+;5'
78'9#$:+;5'
6;51*'
6;
51*'
%#
/13+'/1B+'
/13+'/1B+'
,3&#'
!"#$%&'
!"#$%&'
,2+I1*J312'
+ *J312'
,
2+I1
()#$*+'
()#$*+'
K0
$#B"11B'
K0$#B"11B'
&#
!!>+*51*5'
>+*51*5'
,-.'
,.'
/0+#
+ %&#'
/0+#%&#'
/122)*$3+41*5'
/122)*$3+41*5'
$#
&#
!%#
<$50'=&%>1#?'
<
$50'=&%>1#?'
C)E'F11%'G&&#'
C)E'F11%'G&&#'
C$3#151D'
C$3#151D'
!%+"
!%+"B&5'
+ B&5'
-H3&'<&"1%'
-H3&'<&"1%'
!$#
99+3&A11?'
+3&A11?'
.+/,01,02304*+5#
.+/,01,02304*+5#
.
(+-*+,#6+/#()*+,
(
(+
-*+,#6+/#()*+,
!"#
"#
"#
(+-*+,#
(+-*+,#
'#
/0&&#$15'
/0&&#$15'
(7,01,02304*+5#(+-*+,#
(7,01,02304*+5#(+-*+,#
.+/,01,02304*+5#()*+,
.+/,01,02304*+5#()*+,
Notes: We measured brands’ WOM performance using the log(observed
WOM) minus the expected log(WOM) for the brand based on our model.
The scales are differences in logs (of WOM). Note that in these calculations, we incorporate the category-level random effects but not those of the
brand level.
441
their WOM potential should shape how brands set marketing communications, objectives, and strategies.
The differences between online and offline WOM. We
already know that whereas some brands (e.g., Google, Audi,
eBay) have a strong and active online WOM presence, others (e.g., Coca-Cola, Wal-Mart, Sony) perform well offline.
Our results show that this might be due to differences in
how their characteristics affect WOM in the two channels.
These findings are relevant for managers in at least three
aspects. First, they suggest that copying methods that lead
to success in one medium of communications does not guarantee success in the other. For example, sending samples of
home products to bloggers might not be effective in stimulating them to spread WOM about them.
Second, the findings indicate that following the trend of
relying on measures of online WOM (e.g., NM Incite,
Brandwatch, Netbase, Radian6) to assess success in stimulating a conversation about a brand might not be a good
idea. Our findings imply that such measures might not be
relevant for some managers. Notably, we recently heard a
manager of several well-known household brands express
frustration because she believes that most of the WOM for
her brands comes from offline channels, and yet her performance is measured using online monitoring tools. Our
findings can help to avoid such misaligned incentives. Furthermore, it is worth pointing out that most WOM volume
is still offline (Keller and Fay 2012).
Third, our results call for caution in generalizing findings
from academic studies, because many of these studies were
conducted on online channels—for example, the carryover of
WOM referrals in social networking sites (Trusov, Bucklin,
and Pauwels 2009), the impact of WOM on television viewing (Godes and Mayzlin 2004), and the tone and style of
WOM in blogs (Kozinetz et al. 2010). These findings are valid
for the channels in which they were measured, but generalizing them to offline channels should be done carefully.
A novel benefit of product differentiation. Product differentiation is a key concept in marketing strategy. We advise
brand managers to determine both points of parity and
points of differentiation. To justify the costs of creating differentiation, scholars have attempted to explore their benefits and discuss conditions under which differentiation
should be pursued (Bronnenberg 2008; Dubé 2004;
Schmalensee 1982). Our work contributes to the discussion
by identifying a novel benefit of product differentiation
beyond brand perception and competitive positioning. We
find that differentiated brands have higher WOM and that
this effect is one of the largest among the variables that we
study both online and offline. Our findings could have dramatic implications for managers; for example, it is possible
that even in cases in which differentiation does not have a
direct competitive benefit, its indirect effect through WOM
on sales can justify investment in creating differentiation.
Justifying investment in brands. The large investment in
branding has driven researchers and practitioners to measure the financial outcomes of branding activities and translate brand equity measures into performance metrics such
as profits, customer acquisition, and retention (Leone et al.
2006; Stahl et al. 2012). It is argued that strong brand association leads to enhanced identification and loyalty, which
then translate to higher acquisition and retention rates. Our
findings suggest an additional merit of branding: brand
442
equity has a direct and strong impact on the ability to generate WOM. The four pillars Y&R uses to measure brand
equity as well as the Interbrand top 100 variable play significant roles in explaining WOM. This additional merit
enriches the set of aspects that should be considered for
measuring the impact of brand equity and aids efforts to
reach a more comprehensive understanding of the return on
branding.
LIMITATIONS AND FURTHER RESEARCH
Of course, our study has its limitations. Because we use
cross-sectional, observational data, we cannot empirically
establish a sense of causality. What we can do is examine
both whether the expected effect of each brand characteristic is present after controlling for all other factors and which
effects are most important. As yet, no study has considered
such joint effects for brands and their characteristics on
WOM. Another limitation of our data is that they contain
only the most talked-about brands. As a result, our findings
may not be applicable to brands with relatively little WOM.
Although we have provided some evidence on the robustness of our findings to the selected sample, our data cannot
completely rule out this possibility.
In addition, we relied on measures of aggregate brand
mentions rather than ones disaggregated by source.
Although this aggregation enables us to examine WOM
across many different brands, categories, and channels, a
clearer picture might emerge with regard to mechanisms
underlying specific channel effects through more disaggregate data. Further research could use finer-grained data to
study these and other, more nuanced questions. Along these
lines, this work lays the ground for further research in several directions, outlined in the following subsections.
Channel Effects
In this article, we focused on the relationship between
brand characteristics and WOM and presented results from
online and offline channels as a way to test the generalizability of our findings. However, channel effects convey
many opportunities for further research. More channels can
be explored beyond the offline–online dichotomy. Online
channels (e.g., e-mail, Twitter, blogs, user groups) are different and can show varied patterns of WOM. Gaining a
better understanding of WOM’s dynamics across channels
can help shape strategies for generating WOM, responding
to WOM issues, and identifying leading and lagging indicators of WOM.
Valence and Content
In this study, we counted the overall mentions of WOM,
regardless of other WOM dimensions such as content and
valence. However, brand characteristics may play a role that
goes beyond the mere number of mentions. For valence,
although most WOM is neutral or positive (Keller and Fay
2012), negative WOM is unique because of its possible
impact on adoption and purchase behaviors. Previous studies have explored the implications and contexts of negative
WOM (e.g., Moldovan, Goldenberg, and Chattopadhyay
2011); however, the antecedents of negative WOM have
received little research attention. As for the content, we
present some preliminary evidence indicating that brand
characteristics are also relevant for WOM content, but fur-
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
ther research can shed additional light on characteristics–
content relationships.
Individual-Level Insights
This study examines WOM behaviors at the brand level,
using aggregate measures of WOM. As a result, we cannot
make claims regarding the WOM behaviors of individual
people. For example, do the online–offline variances we
documented result from the same people talking about different brands in different channels, or do different groups
with different interests prefer specific channels? Answering
such questions requires a significantly different and new
data set that tracks the WOM process at the individual level.
To our knowledge, no such data set exists, but building one
could greatly enhance understanding of WOM behaviors at
the individual level.
Moderators of the Brand Characteristics–WOM Relationship
Although this study focuses on the main effects of brand
characteristics, we found in our robustness checks that the
variables related to product type (search, experience, and
credence) may play a more complex role that includes moderation. Future studies could explore such moderating roles.
The goal of the current research is to shed more light on
the intricate relationships between brands and WOM. We
believe that such an understanding can benefit research on
both WOM and brands. The research on WOM will benefit
from understanding the antecedents of WOM, its patterns,
and channel interactions. Branding research will benefit
because WOM is an indicator for market response. This article takes the first step in linking these two literature streams
and providing insight into fruitful areas of further research.
REFERENCES
Aaker, Jennifer L. (1997), “Dimensions of Brand Personality,”
Journal of Marketing Research, 34 (August), 347–56.
Amblee, Naveen and Tung Bui (2008), “Can Brand Reputation
Improve the Odds of Being Reviewed On-Line?” International
Journal of Electronic Commerce, 12 (3), 11–28.
Anand, Bharat N. and Ron Shachar (2011), “Advertising, the
Matchmaker,” RAND Journal of Economics, 42 (2), 205–245.
Anderson, Eugene (1998), “Customer Satisfaction and Word of
Mouth,” Journal of Service Research, 1 (1), 5–17.
Armelini, Guillermo and Julian Villanueva (2010), Marketing
Expenditures and Word-of-Mouth Communication: Complements or Substitutes. Hanover, MA: Now Publishers Inc.
Berg, Julie Dexter, John M. Matthews, and Constance M. O’Hare
(2007), “Measuring Brand Health to Improve Top-Line Growth,”
Sloan Management Review, 49 (1), 61–68.
Berger, Jonah and Chip Heath (2007), “Where Consumers Diverge
from Others: Identity Signaling and Product Domains,” Journal
of Consumer Research, 34 (2), 121–34.
——— and Katherine L. Milkman (2012), “What Makes Online
Content Viral?” Journal of Marketing Research, 49 (April),
192–205.
——— and Eric Schwartz (2011), “What Drives Immediate and
Ongoing Word of Mouth?” Journal of Marketing Research, 48
(October), 869–80.
Bronnenberg, Bart J. (2008), “Brand Competition in CPG Industries: Sustaining Large Local Advantages with Little Product
Differentiation,” Quantitative Marketing and Economics, 6 (1),
79–107.
Che, Hai, Nicholas H. Lurie, and Allen M. Weiss (2011), “Roles,
Incentives, and Contribution Behavior in Online Communities,”
working paper, University of Southern California.
On Brands and Word of Mouth
Cheema, Amar and Andrew M. Kaikati (2010), “The Effect of
Need for Uniqueness on Word of Mouth,” Journal of Marketing
Research, 47 (June), 553–63.
Chevalier, Judith A. and Dina Mayzlin (2006), “The Effect of
Word of Mouth on Sales: Online Book Reviews,” Journal of
Marketing Research, 43 (August), 345–54.
Dichter, Ernest (1966), “How Word-of-Mouth Advertising Works,”
Harvard Business Review, 16 (November/December), 147–66.
Dubé, Jean-Pierre (2004), “Multiple Discreteness and Product Differentiation: Demand for Carbonated Soft Drinks,” Marketing
Science, 23 (1), 66–81.
Fornell, Claes, Michael D. Johnson, Eugene W. Anderson, Jaesung
Cha, and Barbara Everitt Bryant (1996), “The American Customer Satisfaction Index: Nature, Purpose, and Findings,” Journal of Marketing, 60 (October), 7–18.
Godes, David and Dina Mayzlin (2004), “Using Online Conversations to Study Word-of-Mouth Communication,” Marketing Science, 23 (4), 545–60.
Goldenberg, Jacob, Donald Lehmann, Daniella Shidlovski, and
Michal Master Barak (2006), “The Role of Expert Versus Social
Opinion Leader in New Product Adoption,” MSI Working Paper
No. 06-124.
———, Barak Libai, Sarit Moldovan, and Eitan Muller (2007),
“The NPV of Bad News,” International Journal of Research in
Marketing, 24 (3), 186–200.
Han, Young Jee, Joseph C. Nunes, and Xavier Drèze (2010), “Signaling Status with Luxury Goods: The Role of Brand Prominence,” Journal of Marketing, 74 (July), 15–30.
Heath, Chip, Chris Bell, and Emily Sternberg (2001), “Emotional
Selection in Memes: The Case of Urban Legends,” Journal of
Personality and Social Psychology, 81 (6), 1028–41.
Ho, Jason Y.C. and Melanie Dempsey (2010), “Viral Marketing:
Motivations to Forward Online Content,” Journal of Business
Research, 63 (9/10), 1000–1006.
Hoffman, Dona L. and Thomas P. Novak (1996), “Marketing in
Hypermedia Computer-Mediated Environments: Conceptual
Foundations,” Journal of Marketing, 60 (July), 50–68.
Illouz, Eva (2007), Cold Intimacies: The Making of Emotional
Capitalism. Cambridge, UK: Polity Press.
Katona, Zsolt, Peter Zubcsek, and Miklos Sarvary (2011), “Network
Effects and Personal Influences: Diffusion of an Online Social
Network,” Journal of Marketing Research, 48 (June), 425–43.
Keller, Ed (2007), “Unleashing the Power of Word of Mouth: Creating Brand Advocacy to Drive Growth,” Journal of Advertising
Research, 47 (4), 448–52.
——— and Brad Fay (2012), The Face-to-Face Book: Why Real
Relationships Rule in a Digital Marketplace. New York: The
Free Press.
Kozinets, Robert V., Kristine De Valck, Andrea C. Wojnicki, and
Sarah J.S. Wilner (2010), “Networked Narratives: Understanding Word-of-Mouth Marketing in Online Communities,” Journal of Marketing, 74 (March), 71–89.
Laband, David N. (1986), “Advertising as Information: An
Empirical Note,” The Review of Economics and Statistics, 68
(3), 517–52.
Leone, Robert P., Vithala R. Rao, Kevin L. Keller, Anita M. Luo,
Leigh McAlister, and Rajendra Srivastava (2006), “Linking
Brand Equity to Customer Equity,” Journal of Service Research,
9 (2), 125–38.
Lutz, Richard J. and Patrick J. Reilly (1974), “An Exploration of
the Effect of Perceived Social and Performance Risk on Consumer Information Acquisition,” in Advances in Consumer
Research, Vol. 1, Scott Ward and Peter Wright, eds. Ann Arbor,
MI: Association for Consumer Research, 393–405.
Moldovan, Sarit, Jacob Goldenberg, and Amitava Chattopadhyay
(2011), “The Different Roles of Product Originality and Usefulness in Generating Word of Mouth,” International Journal of
Research in Marketing, 28 (2), 109–119.
443
Moore, Gary C. and Izak Benbasat (1991), “Development of an
Instrument to Measure the Perceptions of Adopting an Information Technology Innovation,” Information Systems Research, 2
(3), 192–222.
Morris, Merrill and Christine Ogan (1996), “The Internet as Mass
Medium,” Journal of Computer-Mediated Communication, 1
(4), 39–50.
Mudambi, Susan M. and David Schuff (2010), “What Makes
a Helpful Online Review? A Study of Customer Reviews on
Amazon.com,” MIS Quarterly, 34 (1), 185–200.
Nardi, Bonnie A., Diane J. Schiano, Michelle Gumbrecht, and
Luke Swartz (2004), “Why We Blog,” Communications of the
ACM, 47 (12), 41–46.
Nelson, Phillip (1974), “Advertising as Information,” Journal of
Political Economy, 82 (4), 729–54.
Ostlund, Lyman E. (1974), “Perceived Innovation Attributes as
Predictors of Innovativeness,” Journal of Consumer Research,
1 (2), 23–29.
Parasuraman, A., Valarie A. Zeithaml, and Leonard L. Berry (1985),
“A Conceptual Model of Service Quality and Its Implications for
Future Research,” Journal of Marketing, 49 (October), 41–50.
Peres, Renana, Eitan Muller, and Vijay Mahajan (2010), “Innovation Diffusion and New Product Growth: Critical Review and
Research Directions,” International Journal of Research in
Marketing, 27 (2), 91–106.
Peters, Kim and Yoshihisab Kashima (2007), “From Social Talk to
Social Action: Shaping the Social Triad with Emotion Sharing,”
Journal of Personality and Social Psychology, 93 (5), 780–97.
Power, Daniel J. and Ramesh Sharda (2009), “Decision Support
Systems,” in Springer Handbook of Automation, Shimon Y.
Nof, ed. Berlin: Springer, 1539–48.
Ratchford, Brian T. (1987), “New Insights About the FCB Grid,”
Journal of Advertising Research, 27 (4), 24–38.
Richins, Marsha L. (1983), “Negative Word-of-Mouth by Dissatisfied Consumers: A Pilot Study,” Journal of Marketing, 47 (January), 68–78.
Roberts, Kevin (2004), Lovemarks: The Future Beyond Brands.
New York: PowerHouse Books.
Rogers, Everett M. (1995), The Diffusion of Innovations. New
York: The Free Press.
Rosen, Emanuel (2002), Anatomy of Buzz. New York: Doubleday.
Rubin, Rebecca B., Elizabeth M. Perese, and Carole A. Barbato
(1988), “Conceptualization and Measurement of Interpersonal
Communication Motives,” Human Communication Research,
14 (4), 602–628.
Schafer, Joseph L. and John W. Graham (2002), “Missing Data:
Our View of the State of the Art,” Psychological Methods, 7 (2),
147–77.
Schmalensee, Richard (1982), “Product Differentiation Advantages of Pioneering Brands,” The American Economic Review,
72 (3), 349–65.
Speier, Cheri and Viswanath Venkatesh (2002), “The Hidden
Minefields in the Adoption of Sales Force Automation Technologies,” Journal of Marketing, 66 (July), 98–111.
Stahl, Florian, Mark Heitmann, Donald R. Lehmann, and Scott A.
Neslin (2012), “The Impact of Brand Equity on Customer
Acquisition, Retention, and Profit Margin,” Journal of Marketing, 76 (July), 44–63.
Sundaram, D.S., Kaushik Mitra, and Cynthia Webster (1998),
“Word-of-Mouth Communications: A Motivational Analysis,”
in Advances in Consumer Research, Vol. 25, Joseph W. Alba
and J. Wesley Hutchinson, eds. Provo, UT: Association for Consumer Research, 527–31.
——— and Cynthia Webster (1999), “The Role of Brand Familiarity on the Impact of Word-of-Mouth Communication on
Brand Evaluations,” in Advances in Consumer Research, Vol.
26, Eric J. Arnould and Linda M. Scott, eds. Provo, UT: Association for Consumer Research, 664–70.
444
Trusov, Michael, Randolph E. Bucklin, and Koen Pauwels (2009),
“Effects of Word-of-Mouth Versus Traditional Marketing: Findings from an Internet Social Networking Site,” Journal of Marketing, 73 (September), 90–102.
Veblen, Thorstein (1994[1899]), The Theory of the Leisure Class.
New York: Penguin. First published 1899 by Macmillan Publishers.
Walther, Joseph B. (1996), “Computer-Mediated Communication:
Impersonal, Interpersonal, and Hyperpersonal Interaction,”
Communication Research, 23 (3), 3–43.
Wojnicki, Andrea and David Godes (2011), “Signaling Success:
Strategically Positive Word of Mouth,” working paper, Harvard
Business School.
JOURNAL OF MARKETING RESEARCH, AUGUST 2013
WOMMA (2011), “WOMMA Summit 2012 Announced—Focuses
on Creating Talkable Brands,” press release, (September 29),
(accessed May 14, 2013), [available at http://finance.yahoo.com/
news/WOMMA-Summit-2011-Agenda-iw-2222982791.html].
Yang, Sha, Mantian Hu, Russell S. Winer, Henry Assael, and Xiaohong Chen (2012), “An Empirical Study of Word-of-Mouth
Generation and Consumption,” Marketing Science, 31 (6),
952–63.
Zaichkowsky, Judith L. (1985), “Measuring the Involvement Construct,” Journal of Consumer Research, 12 (3), 341–52.